Making dns_erlang Fast: Binary Tricks for the BEAM

A closer look at Making dns_erlang Fast: Binary Tricks for the BEAM

What this post is about

This is a technical writeup describing how the dns_erlang DNS library was optimised by its authors. It focuses on binary pattern matching, append-friendly accumulators and SWAR tricks for reducing CPU per query on the BEAM. The author explains where the code runs (DNSimple’s cache edge and the authoritative origin running erldns) and why DNS parsing/encoding is hot code: each query through a nameserver goes through dns_erlang twice, once to decode and once to encode.

What you can learn and do from the article

The post teaches concrete BEAM-level ideas: how binaries are represented (small binaries on the process heap, larger binaries as shared off-heap binaries), how sub-binaries and match contexts avoid copies, how append-only accumulators can avoid repeated copying, the integer tagging limits that make small vs bignum values different, and how to inspect binary optimisations with erlc +bin_opt_info or ERL_COMPILER_OPTIONS=bin_opt_info. It also reports empirical results (benchmarked with Benchee on two machines, OTP 29 with the JIT) and explains specific encoder changes, for example replacing many intermediate binaries with a single accumulator that starts with a 12-byte header placeholder.

Who will find this useful

Erlang and Elixir developers who parse or produce binary protocols will get the most from the article, especially teams caring about high-throughput servers. The examples and patterns target people who want to write code that the BEAM compiler and JIT can prove safe to optimise, and who value working through parity tests and benchmarks when changing hot paths.

Questions people ask about Making dns_erlang Fast: Binary Tricks for the BEAM

Does the post show measurable improvements?

Yes; the author reports that library changes reduced CPU per query by 12% and increased queries served per second by 14% on a production-shaped workload, measured with Benchee.

Can I see where binary optimisations apply in my code?

Yes; the post points out you can compile with erlc +bin_opt_info or set ERL_COMPILER_OPTIONS=bin_opt_info to get reports of match contexts and where optimisations are kept or given up.

Are the encoder changes safe and tested?

The author describes running a parity harness that proved byte-for-byte parity with the old code, including error cases, across 26,434 cases for name decoding.

Making dns_erlang Fast: Binary Tricks for the BEAM was shared on NexURLs by Nexis on November 4, 2026, in Tech & AI.